Finance

Pilot wave 2 (M7). Impact: high. See the hub.

Current state (assumptions to validate)

Finance manages the money mechanics of a ~€23.7M-revenue wholesaler running on a net margin around 1.5% — meaning working capital, credit control, and margin discipline are existential, not administrative. Shipping clients pay on account with terms; some (owners/managers abroad) pay through agents, in multi-party chains where the invoice recipient, the vessel, and the payer differ. Defense and public-sector receivables follow their own slow, formal payment cycles. FX exposure exists on imports and some sales. Inventory ties up cash across bonded and free stock.

Assumptions — to validate in discovery: team ~4–6 (distinct from accounting); cash-flow forecasting in Excel from SoftOne exports; receivables chasing by individually written emails; credit limits maintained in SoftOne but reviewed informally; margin reporting after the fact, by period, rarely by order/segment; bank reconciliation partly manual. Pain points: cash-flow forecast effort and staleness, receivables aging with polite-but-slow chasing (English formal correspondence takes time), no early-warning on margin-eroding orders, month-end reporting crunch.

Target AI-first operating model

By M12 finance works from a weekly cash-flow view with AI-drafted commentary explaining movements (what changed, which receipts slipped, which segment drove it), instead of building the spreadsheet each time. Receivables chasing runs as a cadence: an agent drafts stage-appropriate reminders (friendly → firm → escalation) per customer in the right language and tone; the credit controller reviews and sends. Credit-limit reviews are triggered by behavior (aging deterioration, order spikes) rather than calendar. Margin outliers per order/segment surface daily. Humans own every external communication, all credit decisions, and forecasts as judgments — the AI does the assembly and the first-draft narrative.

AI use cases

Use case	Pain addressed	Data needed	Complexity	Impact	Pilot
Receivables-chasing draft emails (staged tone, EN/GR, per-customer context)	Slow, inconsistent chasing on aging AR	SoftOne open items/aging, customer records, payment history	M	H	Y
Cash-flow commentary: weekly narrative on movements and forecast deltas	Manual Excel assembly, stale view	SoftOne AR/AP/bank data, order pipeline	M	H	Y
Credit-limit review briefs triggered by behavior signals	Informal, calendar-based reviews	SoftOne aging, order history, limits	M	M	N
Margin-outlier flagging per order/segment	Margin erosion found late (fatal at 1.5% net)	SoftOne order lines, costs, price lists	M	H	N
Month-end variance narrative (actual vs budget, by segment)	Reporting crunch	SoftOne GL/analytics, budget file	M	M	N
Payment-allocation suggestion for agent-paid multi-vessel remittances	Manual matching of lump payments	Bank statements, open invoices	M	M	N

Process transformation opportunities

• Receivables cadence: from ad hoc chasing to a standing weekly cycle — agent drafts, controller approves, outcomes logged; escalation rules explicit.
• Cash-flow production: from monthly Excel rebuild to an automated weekly refresh + narrative, freeing analysis time.
• Credit control: event-driven reviews with documented briefs, feeding sales before quotes to at-risk accounts (link to sales assistant).
• Margin governance: daily outlier surfacing → weekly margin review with sales/procurement, closing the loop on pricing rules.

Required data sources

• SoftOne: open items/AR aging, AP, GL, order lines with cost/price, customer credit data, bank/cash accounts (module coverage — assumption, to validate).
• Bank statements (export or PSD2 feed); budget/forecast Excel files.
• Finance mailbox (remittance advices, customer payment correspondence).

Potential AI agents

• Receivables Chasing Agent — drafts staged reminder emails per aging bucket; credit controller approves every send.
• Cash-flow Commentary Agent — assembles the weekly view and drafts the narrative; finance manager approves before circulation.
• Margin Watch Agent — flags outlier orders/segments daily; advisory only, no approvals needed (read-only).

Automation opportunities

• n8n: weekly AR-aging extract → chasing-draft batch → controller queue.
• Scheduled daily margin-outlier report; monthly close checklist tracker.
• Bank-statement intake and pre-matching flow.

Required integrations

• SoftOne read (AR/AP, GL, orders, credit data) — see integrations; careful scoping — finance data is the most sensitive read scope in the program, per architecture.
• M365 Outlook (drafts into controller’s mailbox), Excel/SharePoint (budget files).
• Bank statement export (format per bank; PSD2 API optional, year-2).

KPIs

Aligned with the business layer of the KPI framework.

KPI	Baseline	M12 target
DSO (days sales outstanding)	TBD (M2)	−10 days
Overdue AR >60 days (share of total AR)	TBD	−30%
Cash-flow forecast production time	TBD	−70%
Margin outliers reviewed within 48h	0	100%
Chasing drafts sent with minor/no edits	TBD	≥70%

Risks

• Wrong or tone-deaf chasing email damages a key shipping relationship → per-customer tone profiles; controller approval mandatory; key accounts excluded from batch drafting.
• Financial data exposure through the assistant → strictest role scoping in the program; finance corpus segregated; access audited.
• Commentary hallucinating causes for movements → every narrative claim linked to the underlying figures; manager review before circulation.
• Over-trust in forecast narrative → forecasts labeled as drafts; finance manager owns the number.

Training needs

• Credit controller: chasing workflow, approval discipline (M6–M7).
• Finance manager/analysts: commentary review, NL querying of finance data with BI (M8).
• CFO-level: KPI dashboard, margin-review routine (M8).

Deliverables

• Receivables cadence + Chasing Agent (pilot M7, production M9).
• Automated weekly cash-flow pack with commentary.
• Margin Watch reporting; finance KPI dashboard.

12-month execution milestones

Month	Milestone
M1–M2	Discovery: AR process, forecast build walkthrough, data-scope definition, baselines
M3	Finance data-access scoping agreed (security review)
M4	Assistant onboarding (general use only, no finance corpus yet)
M6	Finance read integration tested; controller training
M7	Wave-2 pilot: receivables chasing drafts + cash-flow commentary
M8	Pilot evaluation; margin-outlier reporting live
M9	Agents in production; weekly cadence institutionalized
M11	Credit-review briefs live; month-end narrative pilot
M12	KPI review (DSO, overdue AR); handover